2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
8 * Alan Cox : Fixed the worst of the load
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
36 * The functions in this file will not compile correctly with gcc 2.4.x
39 #include <linux/module.h>
40 #include <linux/types.h>
41 #include <linux/kernel.h>
42 #include <linux/kmemcheck.h>
44 #include <linux/interrupt.h>
46 #include <linux/inet.h>
47 #include <linux/slab.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
62 #include <net/protocol.h>
65 #include <net/checksum.h>
68 #include <asm/uaccess.h>
69 #include <trace/events/skb.h>
70 #include <linux/highmem.h>
72 struct kmem_cache
*skbuff_head_cache __read_mostly
;
73 static struct kmem_cache
*skbuff_fclone_cache __read_mostly
;
75 static void sock_pipe_buf_release(struct pipe_inode_info
*pipe
,
76 struct pipe_buffer
*buf
)
81 static void sock_pipe_buf_get(struct pipe_inode_info
*pipe
,
82 struct pipe_buffer
*buf
)
87 static int sock_pipe_buf_steal(struct pipe_inode_info
*pipe
,
88 struct pipe_buffer
*buf
)
94 /* Pipe buffer operations for a socket. */
95 static const struct pipe_buf_operations sock_pipe_buf_ops
= {
97 .map
= generic_pipe_buf_map
,
98 .unmap
= generic_pipe_buf_unmap
,
99 .confirm
= generic_pipe_buf_confirm
,
100 .release
= sock_pipe_buf_release
,
101 .steal
= sock_pipe_buf_steal
,
102 .get
= sock_pipe_buf_get
,
106 * Keep out-of-line to prevent kernel bloat.
107 * __builtin_return_address is not used because it is not always
112 * skb_over_panic - private function
117 * Out of line support code for skb_put(). Not user callable.
119 static void skb_over_panic(struct sk_buff
*skb
, int sz
, void *here
)
121 printk(KERN_EMERG
"skb_over_panic: text:%p len:%d put:%d head:%p "
122 "data:%p tail:%#lx end:%#lx dev:%s\n",
123 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
124 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
125 skb
->dev
? skb
->dev
->name
: "<NULL>");
130 * skb_under_panic - private function
135 * Out of line support code for skb_push(). Not user callable.
138 static void skb_under_panic(struct sk_buff
*skb
, int sz
, void *here
)
140 printk(KERN_EMERG
"skb_under_panic: text:%p len:%d put:%d head:%p "
141 "data:%p tail:%#lx end:%#lx dev:%s\n",
142 here
, skb
->len
, sz
, skb
->head
, skb
->data
,
143 (unsigned long)skb
->tail
, (unsigned long)skb
->end
,
144 skb
->dev
? skb
->dev
->name
: "<NULL>");
148 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 struct sk_buff
*__alloc_skb(unsigned int size
, gfp_t gfp_mask
,
170 int fclone
, int node
)
172 struct kmem_cache
*cache
;
173 struct skb_shared_info
*shinfo
;
177 cache
= fclone
? skbuff_fclone_cache
: skbuff_head_cache
;
180 skb
= kmem_cache_alloc_node(cache
, gfp_mask
& ~__GFP_DMA
, node
);
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
190 size
= SKB_DATA_ALIGN(size
);
191 size
+= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
192 data
= kmalloc_node_track_caller(size
, gfp_mask
, node
);
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
199 size
= SKB_WITH_OVERHEAD(ksize(data
));
200 prefetchw(data
+ size
);
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
207 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
208 /* Account for allocated memory : skb + skb->head */
209 skb
->truesize
= SKB_TRUESIZE(size
);
210 atomic_set(&skb
->users
, 1);
213 skb_reset_tail_pointer(skb
);
214 skb
->end
= skb
->tail
+ size
;
215 #ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb
->mac_header
= ~0U;
219 /* make sure we initialize shinfo sequentially */
220 shinfo
= skb_shinfo(skb
);
221 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
222 atomic_set(&shinfo
->dataref
, 1);
223 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
226 struct sk_buff
*child
= skb
+ 1;
227 atomic_t
*fclone_ref
= (atomic_t
*) (child
+ 1);
229 kmemcheck_annotate_bitfield(child
, flags1
);
230 kmemcheck_annotate_bitfield(child
, flags2
);
231 skb
->fclone
= SKB_FCLONE_ORIG
;
232 atomic_set(fclone_ref
, 1);
234 child
->fclone
= SKB_FCLONE_UNAVAILABLE
;
239 kmem_cache_free(cache
, skb
);
243 EXPORT_SYMBOL(__alloc_skb
);
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
248 * @frag_size: size of fragment, or 0 if head was kmalloced
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
262 struct sk_buff
*build_skb(void *data
, unsigned int frag_size
)
264 struct skb_shared_info
*shinfo
;
266 unsigned int size
= frag_size
? : ksize(data
);
268 skb
= kmem_cache_alloc(skbuff_head_cache
, GFP_ATOMIC
);
272 size
-= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
274 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
275 skb
->truesize
= SKB_TRUESIZE(size
);
276 skb
->head_frag
= frag_size
!= 0;
277 atomic_set(&skb
->users
, 1);
280 skb_reset_tail_pointer(skb
);
281 skb
->end
= skb
->tail
+ size
;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb
->mac_header
= ~0U;
286 /* make sure we initialize shinfo sequentially */
287 shinfo
= skb_shinfo(skb
);
288 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
289 atomic_set(&shinfo
->dataref
, 1);
290 kmemcheck_annotate_variable(shinfo
->destructor_arg
);
294 EXPORT_SYMBOL(build_skb
);
297 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
298 * @dev: network device to receive on
299 * @length: length to allocate
300 * @gfp_mask: get_free_pages mask, passed to alloc_skb
302 * Allocate a new &sk_buff and assign it a usage count of one. The
303 * buffer has unspecified headroom built in. Users should allocate
304 * the headroom they think they need without accounting for the
305 * built in space. The built in space is used for optimisations.
307 * %NULL is returned if there is no free memory.
309 struct sk_buff
*__netdev_alloc_skb(struct net_device
*dev
,
310 unsigned int length
, gfp_t gfp_mask
)
314 skb
= __alloc_skb(length
+ NET_SKB_PAD
, gfp_mask
, 0, NUMA_NO_NODE
);
316 skb_reserve(skb
, NET_SKB_PAD
);
321 EXPORT_SYMBOL(__netdev_alloc_skb
);
323 void skb_add_rx_frag(struct sk_buff
*skb
, int i
, struct page
*page
, int off
,
324 int size
, unsigned int truesize
)
326 skb_fill_page_desc(skb
, i
, page
, off
, size
);
328 skb
->data_len
+= size
;
329 skb
->truesize
+= truesize
;
331 EXPORT_SYMBOL(skb_add_rx_frag
);
334 * dev_alloc_skb - allocate an skbuff for receiving
335 * @length: length to allocate
337 * Allocate a new &sk_buff and assign it a usage count of one. The
338 * buffer has unspecified headroom built in. Users should allocate
339 * the headroom they think they need without accounting for the
340 * built in space. The built in space is used for optimisations.
342 * %NULL is returned if there is no free memory. Although this function
343 * allocates memory it can be called from an interrupt.
345 struct sk_buff
*dev_alloc_skb(unsigned int length
)
348 * There is more code here than it seems:
349 * __dev_alloc_skb is an inline
351 return __dev_alloc_skb(length
, GFP_ATOMIC
);
353 EXPORT_SYMBOL(dev_alloc_skb
);
355 static void skb_drop_list(struct sk_buff
**listp
)
357 struct sk_buff
*list
= *listp
;
362 struct sk_buff
*this = list
;
368 static inline void skb_drop_fraglist(struct sk_buff
*skb
)
370 skb_drop_list(&skb_shinfo(skb
)->frag_list
);
373 static void skb_clone_fraglist(struct sk_buff
*skb
)
375 struct sk_buff
*list
;
377 skb_walk_frags(skb
, list
)
381 static void skb_free_head(struct sk_buff
*skb
)
384 put_page(virt_to_head_page(skb
->head
));
389 static void skb_release_data(struct sk_buff
*skb
)
392 !atomic_sub_return(skb
->nohdr
? (1 << SKB_DATAREF_SHIFT
) + 1 : 1,
393 &skb_shinfo(skb
)->dataref
)) {
394 if (skb_shinfo(skb
)->nr_frags
) {
396 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
397 skb_frag_unref(skb
, i
);
401 * If skb buf is from userspace, we need to notify the caller
402 * the lower device DMA has done;
404 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
405 struct ubuf_info
*uarg
;
407 uarg
= skb_shinfo(skb
)->destructor_arg
;
409 uarg
->callback(uarg
);
412 if (skb_has_frag_list(skb
))
413 skb_drop_fraglist(skb
);
420 * Free an skbuff by memory without cleaning the state.
422 static void kfree_skbmem(struct sk_buff
*skb
)
424 struct sk_buff
*other
;
425 atomic_t
*fclone_ref
;
427 switch (skb
->fclone
) {
428 case SKB_FCLONE_UNAVAILABLE
:
429 kmem_cache_free(skbuff_head_cache
, skb
);
432 case SKB_FCLONE_ORIG
:
433 fclone_ref
= (atomic_t
*) (skb
+ 2);
434 if (atomic_dec_and_test(fclone_ref
))
435 kmem_cache_free(skbuff_fclone_cache
, skb
);
438 case SKB_FCLONE_CLONE
:
439 fclone_ref
= (atomic_t
*) (skb
+ 1);
442 /* The clone portion is available for
443 * fast-cloning again.
445 skb
->fclone
= SKB_FCLONE_UNAVAILABLE
;
447 if (atomic_dec_and_test(fclone_ref
))
448 kmem_cache_free(skbuff_fclone_cache
, other
);
453 static void skb_release_head_state(struct sk_buff
*skb
)
457 secpath_put(skb
->sp
);
459 if (skb
->destructor
) {
461 skb
->destructor(skb
);
463 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
464 nf_conntrack_put(skb
->nfct
);
466 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
467 nf_conntrack_put_reasm(skb
->nfct_reasm
);
469 #ifdef CONFIG_BRIDGE_NETFILTER
470 nf_bridge_put(skb
->nf_bridge
);
472 /* XXX: IS this still necessary? - JHS */
473 #ifdef CONFIG_NET_SCHED
475 #ifdef CONFIG_NET_CLS_ACT
481 /* Free everything but the sk_buff shell. */
482 static void skb_release_all(struct sk_buff
*skb
)
484 skb_release_head_state(skb
);
485 skb_release_data(skb
);
489 * __kfree_skb - private function
492 * Free an sk_buff. Release anything attached to the buffer.
493 * Clean the state. This is an internal helper function. Users should
494 * always call kfree_skb
497 void __kfree_skb(struct sk_buff
*skb
)
499 skb_release_all(skb
);
502 EXPORT_SYMBOL(__kfree_skb
);
505 * kfree_skb - free an sk_buff
506 * @skb: buffer to free
508 * Drop a reference to the buffer and free it if the usage count has
511 void kfree_skb(struct sk_buff
*skb
)
515 if (likely(atomic_read(&skb
->users
) == 1))
517 else if (likely(!atomic_dec_and_test(&skb
->users
)))
519 trace_kfree_skb(skb
, __builtin_return_address(0));
522 EXPORT_SYMBOL(kfree_skb
);
525 * consume_skb - free an skbuff
526 * @skb: buffer to free
528 * Drop a ref to the buffer and free it if the usage count has hit zero
529 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
530 * is being dropped after a failure and notes that
532 void consume_skb(struct sk_buff
*skb
)
536 if (likely(atomic_read(&skb
->users
) == 1))
538 else if (likely(!atomic_dec_and_test(&skb
->users
)))
540 trace_consume_skb(skb
);
543 EXPORT_SYMBOL(consume_skb
);
546 * skb_recycle - clean up an skb for reuse
549 * Recycles the skb to be reused as a receive buffer. This
550 * function does any necessary reference count dropping, and
551 * cleans up the skbuff as if it just came from __alloc_skb().
553 void skb_recycle(struct sk_buff
*skb
)
555 struct skb_shared_info
*shinfo
;
557 skb_release_head_state(skb
);
559 shinfo
= skb_shinfo(skb
);
560 memset(shinfo
, 0, offsetof(struct skb_shared_info
, dataref
));
561 atomic_set(&shinfo
->dataref
, 1);
563 memset(skb
, 0, offsetof(struct sk_buff
, tail
));
564 skb
->data
= skb
->head
+ NET_SKB_PAD
;
565 skb_reset_tail_pointer(skb
);
567 EXPORT_SYMBOL(skb_recycle
);
570 * skb_recycle_check - check if skb can be reused for receive
572 * @skb_size: minimum receive buffer size
574 * Checks that the skb passed in is not shared or cloned, and
575 * that it is linear and its head portion at least as large as
576 * skb_size so that it can be recycled as a receive buffer.
577 * If these conditions are met, this function does any necessary
578 * reference count dropping and cleans up the skbuff as if it
579 * just came from __alloc_skb().
581 bool skb_recycle_check(struct sk_buff
*skb
, int skb_size
)
583 if (!skb_is_recycleable(skb
, skb_size
))
590 EXPORT_SYMBOL(skb_recycle_check
);
592 static void __copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
594 new->tstamp
= old
->tstamp
;
596 new->transport_header
= old
->transport_header
;
597 new->network_header
= old
->network_header
;
598 new->mac_header
= old
->mac_header
;
599 skb_dst_copy(new, old
);
600 new->rxhash
= old
->rxhash
;
601 new->ooo_okay
= old
->ooo_okay
;
602 new->l4_rxhash
= old
->l4_rxhash
;
603 new->no_fcs
= old
->no_fcs
;
605 new->sp
= secpath_get(old
->sp
);
607 memcpy(new->cb
, old
->cb
, sizeof(old
->cb
));
608 new->csum
= old
->csum
;
609 new->local_df
= old
->local_df
;
610 new->pkt_type
= old
->pkt_type
;
611 new->ip_summed
= old
->ip_summed
;
612 skb_copy_queue_mapping(new, old
);
613 new->priority
= old
->priority
;
614 #if IS_ENABLED(CONFIG_IP_VS)
615 new->ipvs_property
= old
->ipvs_property
;
617 new->protocol
= old
->protocol
;
618 new->mark
= old
->mark
;
619 new->skb_iif
= old
->skb_iif
;
621 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
622 new->nf_trace
= old
->nf_trace
;
624 #ifdef CONFIG_NET_SCHED
625 new->tc_index
= old
->tc_index
;
626 #ifdef CONFIG_NET_CLS_ACT
627 new->tc_verd
= old
->tc_verd
;
630 new->vlan_tci
= old
->vlan_tci
;
632 skb_copy_secmark(new, old
);
636 * You should not add any new code to this function. Add it to
637 * __copy_skb_header above instead.
639 static struct sk_buff
*__skb_clone(struct sk_buff
*n
, struct sk_buff
*skb
)
641 #define C(x) n->x = skb->x
643 n
->next
= n
->prev
= NULL
;
645 __copy_skb_header(n
, skb
);
650 n
->hdr_len
= skb
->nohdr
? skb_headroom(skb
) : skb
->hdr_len
;
653 n
->destructor
= NULL
;
660 atomic_set(&n
->users
, 1);
662 atomic_inc(&(skb_shinfo(skb
)->dataref
));
670 * skb_morph - morph one skb into another
671 * @dst: the skb to receive the contents
672 * @src: the skb to supply the contents
674 * This is identical to skb_clone except that the target skb is
675 * supplied by the user.
677 * The target skb is returned upon exit.
679 struct sk_buff
*skb_morph(struct sk_buff
*dst
, struct sk_buff
*src
)
681 skb_release_all(dst
);
682 return __skb_clone(dst
, src
);
684 EXPORT_SYMBOL_GPL(skb_morph
);
686 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
687 * @skb: the skb to modify
688 * @gfp_mask: allocation priority
690 * This must be called on SKBTX_DEV_ZEROCOPY skb.
691 * It will copy all frags into kernel and drop the reference
692 * to userspace pages.
694 * If this function is called from an interrupt gfp_mask() must be
697 * Returns 0 on success or a negative error code on failure
698 * to allocate kernel memory to copy to.
700 int skb_copy_ubufs(struct sk_buff
*skb
, gfp_t gfp_mask
)
703 int num_frags
= skb_shinfo(skb
)->nr_frags
;
704 struct page
*page
, *head
= NULL
;
705 struct ubuf_info
*uarg
= skb_shinfo(skb
)->destructor_arg
;
707 for (i
= 0; i
< num_frags
; i
++) {
709 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
711 page
= alloc_page(GFP_ATOMIC
);
714 struct page
*next
= (struct page
*)head
->private;
720 vaddr
= kmap_atomic(skb_frag_page(f
));
721 memcpy(page_address(page
),
722 vaddr
+ f
->page_offset
, skb_frag_size(f
));
723 kunmap_atomic(vaddr
);
724 page
->private = (unsigned long)head
;
728 /* skb frags release userspace buffers */
729 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
730 skb_frag_unref(skb
, i
);
732 uarg
->callback(uarg
);
734 /* skb frags point to kernel buffers */
735 for (i
= skb_shinfo(skb
)->nr_frags
; i
> 0; i
--) {
736 __skb_fill_page_desc(skb
, i
-1, head
, 0,
737 skb_shinfo(skb
)->frags
[i
- 1].size
);
738 head
= (struct page
*)head
->private;
741 skb_shinfo(skb
)->tx_flags
&= ~SKBTX_DEV_ZEROCOPY
;
747 * skb_clone - duplicate an sk_buff
748 * @skb: buffer to clone
749 * @gfp_mask: allocation priority
751 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
752 * copies share the same packet data but not structure. The new
753 * buffer has a reference count of 1. If the allocation fails the
754 * function returns %NULL otherwise the new buffer is returned.
756 * If this function is called from an interrupt gfp_mask() must be
760 struct sk_buff
*skb_clone(struct sk_buff
*skb
, gfp_t gfp_mask
)
764 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
765 if (skb_copy_ubufs(skb
, gfp_mask
))
770 if (skb
->fclone
== SKB_FCLONE_ORIG
&&
771 n
->fclone
== SKB_FCLONE_UNAVAILABLE
) {
772 atomic_t
*fclone_ref
= (atomic_t
*) (n
+ 1);
773 n
->fclone
= SKB_FCLONE_CLONE
;
774 atomic_inc(fclone_ref
);
776 n
= kmem_cache_alloc(skbuff_head_cache
, gfp_mask
);
780 kmemcheck_annotate_bitfield(n
, flags1
);
781 kmemcheck_annotate_bitfield(n
, flags2
);
782 n
->fclone
= SKB_FCLONE_UNAVAILABLE
;
785 return __skb_clone(n
, skb
);
787 EXPORT_SYMBOL(skb_clone
);
789 static void copy_skb_header(struct sk_buff
*new, const struct sk_buff
*old
)
791 #ifndef NET_SKBUFF_DATA_USES_OFFSET
793 * Shift between the two data areas in bytes
795 unsigned long offset
= new->data
- old
->data
;
798 __copy_skb_header(new, old
);
800 #ifndef NET_SKBUFF_DATA_USES_OFFSET
801 /* {transport,network,mac}_header are relative to skb->head */
802 new->transport_header
+= offset
;
803 new->network_header
+= offset
;
804 if (skb_mac_header_was_set(new))
805 new->mac_header
+= offset
;
807 skb_shinfo(new)->gso_size
= skb_shinfo(old
)->gso_size
;
808 skb_shinfo(new)->gso_segs
= skb_shinfo(old
)->gso_segs
;
809 skb_shinfo(new)->gso_type
= skb_shinfo(old
)->gso_type
;
813 * skb_copy - create private copy of an sk_buff
814 * @skb: buffer to copy
815 * @gfp_mask: allocation priority
817 * Make a copy of both an &sk_buff and its data. This is used when the
818 * caller wishes to modify the data and needs a private copy of the
819 * data to alter. Returns %NULL on failure or the pointer to the buffer
820 * on success. The returned buffer has a reference count of 1.
822 * As by-product this function converts non-linear &sk_buff to linear
823 * one, so that &sk_buff becomes completely private and caller is allowed
824 * to modify all the data of returned buffer. This means that this
825 * function is not recommended for use in circumstances when only
826 * header is going to be modified. Use pskb_copy() instead.
829 struct sk_buff
*skb_copy(const struct sk_buff
*skb
, gfp_t gfp_mask
)
831 int headerlen
= skb_headroom(skb
);
832 unsigned int size
= skb_end_offset(skb
) + skb
->data_len
;
833 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
838 /* Set the data pointer */
839 skb_reserve(n
, headerlen
);
840 /* Set the tail pointer and length */
841 skb_put(n
, skb
->len
);
843 if (skb_copy_bits(skb
, -headerlen
, n
->head
, headerlen
+ skb
->len
))
846 copy_skb_header(n
, skb
);
849 EXPORT_SYMBOL(skb_copy
);
852 * __pskb_copy - create copy of an sk_buff with private head.
853 * @skb: buffer to copy
854 * @headroom: headroom of new skb
855 * @gfp_mask: allocation priority
857 * Make a copy of both an &sk_buff and part of its data, located
858 * in header. Fragmented data remain shared. This is used when
859 * the caller wishes to modify only header of &sk_buff and needs
860 * private copy of the header to alter. Returns %NULL on failure
861 * or the pointer to the buffer on success.
862 * The returned buffer has a reference count of 1.
865 struct sk_buff
*__pskb_copy(struct sk_buff
*skb
, int headroom
, gfp_t gfp_mask
)
867 unsigned int size
= skb_headlen(skb
) + headroom
;
868 struct sk_buff
*n
= alloc_skb(size
, gfp_mask
);
873 /* Set the data pointer */
874 skb_reserve(n
, headroom
);
875 /* Set the tail pointer and length */
876 skb_put(n
, skb_headlen(skb
));
878 skb_copy_from_linear_data(skb
, n
->data
, n
->len
);
880 n
->truesize
+= skb
->data_len
;
881 n
->data_len
= skb
->data_len
;
884 if (skb_shinfo(skb
)->nr_frags
) {
887 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
888 if (skb_copy_ubufs(skb
, gfp_mask
)) {
894 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
895 skb_shinfo(n
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
896 skb_frag_ref(skb
, i
);
898 skb_shinfo(n
)->nr_frags
= i
;
901 if (skb_has_frag_list(skb
)) {
902 skb_shinfo(n
)->frag_list
= skb_shinfo(skb
)->frag_list
;
903 skb_clone_fraglist(n
);
906 copy_skb_header(n
, skb
);
910 EXPORT_SYMBOL(__pskb_copy
);
913 * pskb_expand_head - reallocate header of &sk_buff
914 * @skb: buffer to reallocate
915 * @nhead: room to add at head
916 * @ntail: room to add at tail
917 * @gfp_mask: allocation priority
919 * Expands (or creates identical copy, if &nhead and &ntail are zero)
920 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
921 * reference count of 1. Returns zero in the case of success or error,
922 * if expansion failed. In the last case, &sk_buff is not changed.
924 * All the pointers pointing into skb header may change and must be
925 * reloaded after call to this function.
928 int pskb_expand_head(struct sk_buff
*skb
, int nhead
, int ntail
,
933 int size
= nhead
+ skb_end_offset(skb
) + ntail
;
941 size
= SKB_DATA_ALIGN(size
);
943 data
= kmalloc(size
+ SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)),
947 size
= SKB_WITH_OVERHEAD(ksize(data
));
949 /* Copy only real data... and, alas, header. This should be
950 * optimized for the cases when header is void.
952 memcpy(data
+ nhead
, skb
->head
, skb_tail_pointer(skb
) - skb
->head
);
954 memcpy((struct skb_shared_info
*)(data
+ size
),
956 offsetof(struct skb_shared_info
, frags
[skb_shinfo(skb
)->nr_frags
]));
959 * if shinfo is shared we must drop the old head gracefully, but if it
960 * is not we can just drop the old head and let the existing refcount
961 * be since all we did is relocate the values
963 if (skb_cloned(skb
)) {
964 /* copy this zero copy skb frags */
965 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
966 if (skb_copy_ubufs(skb
, gfp_mask
))
969 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
970 skb_frag_ref(skb
, i
);
972 if (skb_has_frag_list(skb
))
973 skb_clone_fraglist(skb
);
975 skb_release_data(skb
);
979 off
= (data
+ nhead
) - skb
->head
;
984 #ifdef NET_SKBUFF_DATA_USES_OFFSET
988 skb
->end
= skb
->head
+ size
;
990 /* {transport,network,mac}_header and tail are relative to skb->head */
992 skb
->transport_header
+= off
;
993 skb
->network_header
+= off
;
994 if (skb_mac_header_was_set(skb
))
995 skb
->mac_header
+= off
;
996 /* Only adjust this if it actually is csum_start rather than csum */
997 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
998 skb
->csum_start
+= nhead
;
1002 atomic_set(&skb_shinfo(skb
)->dataref
, 1);
1010 EXPORT_SYMBOL(pskb_expand_head
);
1012 /* Make private copy of skb with writable head and some headroom */
1014 struct sk_buff
*skb_realloc_headroom(struct sk_buff
*skb
, unsigned int headroom
)
1016 struct sk_buff
*skb2
;
1017 int delta
= headroom
- skb_headroom(skb
);
1020 skb2
= pskb_copy(skb
, GFP_ATOMIC
);
1022 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1023 if (skb2
&& pskb_expand_head(skb2
, SKB_DATA_ALIGN(delta
), 0,
1031 EXPORT_SYMBOL(skb_realloc_headroom
);
1034 * skb_copy_expand - copy and expand sk_buff
1035 * @skb: buffer to copy
1036 * @newheadroom: new free bytes at head
1037 * @newtailroom: new free bytes at tail
1038 * @gfp_mask: allocation priority
1040 * Make a copy of both an &sk_buff and its data and while doing so
1041 * allocate additional space.
1043 * This is used when the caller wishes to modify the data and needs a
1044 * private copy of the data to alter as well as more space for new fields.
1045 * Returns %NULL on failure or the pointer to the buffer
1046 * on success. The returned buffer has a reference count of 1.
1048 * You must pass %GFP_ATOMIC as the allocation priority if this function
1049 * is called from an interrupt.
1051 struct sk_buff
*skb_copy_expand(const struct sk_buff
*skb
,
1052 int newheadroom
, int newtailroom
,
1056 * Allocate the copy buffer
1058 struct sk_buff
*n
= alloc_skb(newheadroom
+ skb
->len
+ newtailroom
,
1060 int oldheadroom
= skb_headroom(skb
);
1061 int head_copy_len
, head_copy_off
;
1067 skb_reserve(n
, newheadroom
);
1069 /* Set the tail pointer and length */
1070 skb_put(n
, skb
->len
);
1072 head_copy_len
= oldheadroom
;
1074 if (newheadroom
<= head_copy_len
)
1075 head_copy_len
= newheadroom
;
1077 head_copy_off
= newheadroom
- head_copy_len
;
1079 /* Copy the linear header and data. */
1080 if (skb_copy_bits(skb
, -head_copy_len
, n
->head
+ head_copy_off
,
1081 skb
->len
+ head_copy_len
))
1084 copy_skb_header(n
, skb
);
1086 off
= newheadroom
- oldheadroom
;
1087 if (n
->ip_summed
== CHECKSUM_PARTIAL
)
1088 n
->csum_start
+= off
;
1089 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1090 n
->transport_header
+= off
;
1091 n
->network_header
+= off
;
1092 if (skb_mac_header_was_set(skb
))
1093 n
->mac_header
+= off
;
1098 EXPORT_SYMBOL(skb_copy_expand
);
1101 * skb_pad - zero pad the tail of an skb
1102 * @skb: buffer to pad
1103 * @pad: space to pad
1105 * Ensure that a buffer is followed by a padding area that is zero
1106 * filled. Used by network drivers which may DMA or transfer data
1107 * beyond the buffer end onto the wire.
1109 * May return error in out of memory cases. The skb is freed on error.
1112 int skb_pad(struct sk_buff
*skb
, int pad
)
1117 /* If the skbuff is non linear tailroom is always zero.. */
1118 if (!skb_cloned(skb
) && skb_tailroom(skb
) >= pad
) {
1119 memset(skb
->data
+skb
->len
, 0, pad
);
1123 ntail
= skb
->data_len
+ pad
- (skb
->end
- skb
->tail
);
1124 if (likely(skb_cloned(skb
) || ntail
> 0)) {
1125 err
= pskb_expand_head(skb
, 0, ntail
, GFP_ATOMIC
);
1130 /* FIXME: The use of this function with non-linear skb's really needs
1133 err
= skb_linearize(skb
);
1137 memset(skb
->data
+ skb
->len
, 0, pad
);
1144 EXPORT_SYMBOL(skb_pad
);
1147 * skb_put - add data to a buffer
1148 * @skb: buffer to use
1149 * @len: amount of data to add
1151 * This function extends the used data area of the buffer. If this would
1152 * exceed the total buffer size the kernel will panic. A pointer to the
1153 * first byte of the extra data is returned.
1155 unsigned char *skb_put(struct sk_buff
*skb
, unsigned int len
)
1157 unsigned char *tmp
= skb_tail_pointer(skb
);
1158 SKB_LINEAR_ASSERT(skb
);
1161 if (unlikely(skb
->tail
> skb
->end
))
1162 skb_over_panic(skb
, len
, __builtin_return_address(0));
1165 EXPORT_SYMBOL(skb_put
);
1168 * skb_push - add data to the start of a buffer
1169 * @skb: buffer to use
1170 * @len: amount of data to add
1172 * This function extends the used data area of the buffer at the buffer
1173 * start. If this would exceed the total buffer headroom the kernel will
1174 * panic. A pointer to the first byte of the extra data is returned.
1176 unsigned char *skb_push(struct sk_buff
*skb
, unsigned int len
)
1180 if (unlikely(skb
->data
<skb
->head
))
1181 skb_under_panic(skb
, len
, __builtin_return_address(0));
1184 EXPORT_SYMBOL(skb_push
);
1187 * skb_pull - remove data from the start of a buffer
1188 * @skb: buffer to use
1189 * @len: amount of data to remove
1191 * This function removes data from the start of a buffer, returning
1192 * the memory to the headroom. A pointer to the next data in the buffer
1193 * is returned. Once the data has been pulled future pushes will overwrite
1196 unsigned char *skb_pull(struct sk_buff
*skb
, unsigned int len
)
1198 return skb_pull_inline(skb
, len
);
1200 EXPORT_SYMBOL(skb_pull
);
1203 * skb_trim - remove end from a buffer
1204 * @skb: buffer to alter
1207 * Cut the length of a buffer down by removing data from the tail. If
1208 * the buffer is already under the length specified it is not modified.
1209 * The skb must be linear.
1211 void skb_trim(struct sk_buff
*skb
, unsigned int len
)
1214 __skb_trim(skb
, len
);
1216 EXPORT_SYMBOL(skb_trim
);
1218 /* Trims skb to length len. It can change skb pointers.
1221 int ___pskb_trim(struct sk_buff
*skb
, unsigned int len
)
1223 struct sk_buff
**fragp
;
1224 struct sk_buff
*frag
;
1225 int offset
= skb_headlen(skb
);
1226 int nfrags
= skb_shinfo(skb
)->nr_frags
;
1230 if (skb_cloned(skb
) &&
1231 unlikely((err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
))))
1238 for (; i
< nfrags
; i
++) {
1239 int end
= offset
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1246 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
++], len
- offset
);
1249 skb_shinfo(skb
)->nr_frags
= i
;
1251 for (; i
< nfrags
; i
++)
1252 skb_frag_unref(skb
, i
);
1254 if (skb_has_frag_list(skb
))
1255 skb_drop_fraglist(skb
);
1259 for (fragp
= &skb_shinfo(skb
)->frag_list
; (frag
= *fragp
);
1260 fragp
= &frag
->next
) {
1261 int end
= offset
+ frag
->len
;
1263 if (skb_shared(frag
)) {
1264 struct sk_buff
*nfrag
;
1266 nfrag
= skb_clone(frag
, GFP_ATOMIC
);
1267 if (unlikely(!nfrag
))
1270 nfrag
->next
= frag
->next
;
1282 unlikely((err
= pskb_trim(frag
, len
- offset
))))
1286 skb_drop_list(&frag
->next
);
1291 if (len
> skb_headlen(skb
)) {
1292 skb
->data_len
-= skb
->len
- len
;
1297 skb_set_tail_pointer(skb
, len
);
1302 EXPORT_SYMBOL(___pskb_trim
);
1305 * __pskb_pull_tail - advance tail of skb header
1306 * @skb: buffer to reallocate
1307 * @delta: number of bytes to advance tail
1309 * The function makes a sense only on a fragmented &sk_buff,
1310 * it expands header moving its tail forward and copying necessary
1311 * data from fragmented part.
1313 * &sk_buff MUST have reference count of 1.
1315 * Returns %NULL (and &sk_buff does not change) if pull failed
1316 * or value of new tail of skb in the case of success.
1318 * All the pointers pointing into skb header may change and must be
1319 * reloaded after call to this function.
1322 /* Moves tail of skb head forward, copying data from fragmented part,
1323 * when it is necessary.
1324 * 1. It may fail due to malloc failure.
1325 * 2. It may change skb pointers.
1327 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1329 unsigned char *__pskb_pull_tail(struct sk_buff
*skb
, int delta
)
1331 /* If skb has not enough free space at tail, get new one
1332 * plus 128 bytes for future expansions. If we have enough
1333 * room at tail, reallocate without expansion only if skb is cloned.
1335 int i
, k
, eat
= (skb
->tail
+ delta
) - skb
->end
;
1337 if (eat
> 0 || skb_cloned(skb
)) {
1338 if (pskb_expand_head(skb
, 0, eat
> 0 ? eat
+ 128 : 0,
1343 if (skb_copy_bits(skb
, skb_headlen(skb
), skb_tail_pointer(skb
), delta
))
1346 /* Optimization: no fragments, no reasons to preestimate
1347 * size of pulled pages. Superb.
1349 if (!skb_has_frag_list(skb
))
1352 /* Estimate size of pulled pages. */
1354 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1355 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1362 /* If we need update frag list, we are in troubles.
1363 * Certainly, it possible to add an offset to skb data,
1364 * but taking into account that pulling is expected to
1365 * be very rare operation, it is worth to fight against
1366 * further bloating skb head and crucify ourselves here instead.
1367 * Pure masohism, indeed. 8)8)
1370 struct sk_buff
*list
= skb_shinfo(skb
)->frag_list
;
1371 struct sk_buff
*clone
= NULL
;
1372 struct sk_buff
*insp
= NULL
;
1377 if (list
->len
<= eat
) {
1378 /* Eaten as whole. */
1383 /* Eaten partially. */
1385 if (skb_shared(list
)) {
1386 /* Sucks! We need to fork list. :-( */
1387 clone
= skb_clone(list
, GFP_ATOMIC
);
1393 /* This may be pulled without
1397 if (!pskb_pull(list
, eat
)) {
1405 /* Free pulled out fragments. */
1406 while ((list
= skb_shinfo(skb
)->frag_list
) != insp
) {
1407 skb_shinfo(skb
)->frag_list
= list
->next
;
1410 /* And insert new clone at head. */
1413 skb_shinfo(skb
)->frag_list
= clone
;
1416 /* Success! Now we may commit changes to skb data. */
1421 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1422 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1425 skb_frag_unref(skb
, i
);
1428 skb_shinfo(skb
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
1430 skb_shinfo(skb
)->frags
[k
].page_offset
+= eat
;
1431 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[k
], eat
);
1437 skb_shinfo(skb
)->nr_frags
= k
;
1440 skb
->data_len
-= delta
;
1442 return skb_tail_pointer(skb
);
1444 EXPORT_SYMBOL(__pskb_pull_tail
);
1447 * skb_copy_bits - copy bits from skb to kernel buffer
1449 * @offset: offset in source
1450 * @to: destination buffer
1451 * @len: number of bytes to copy
1453 * Copy the specified number of bytes from the source skb to the
1454 * destination buffer.
1457 * If its prototype is ever changed,
1458 * check arch/{*}/net/{*}.S files,
1459 * since it is called from BPF assembly code.
1461 int skb_copy_bits(const struct sk_buff
*skb
, int offset
, void *to
, int len
)
1463 int start
= skb_headlen(skb
);
1464 struct sk_buff
*frag_iter
;
1467 if (offset
> (int)skb
->len
- len
)
1471 if ((copy
= start
- offset
) > 0) {
1474 skb_copy_from_linear_data_offset(skb
, offset
, to
, copy
);
1475 if ((len
-= copy
) == 0)
1481 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1483 skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[i
];
1485 WARN_ON(start
> offset
+ len
);
1487 end
= start
+ skb_frag_size(f
);
1488 if ((copy
= end
- offset
) > 0) {
1494 vaddr
= kmap_atomic(skb_frag_page(f
));
1496 vaddr
+ f
->page_offset
+ offset
- start
,
1498 kunmap_atomic(vaddr
);
1500 if ((len
-= copy
) == 0)
1508 skb_walk_frags(skb
, frag_iter
) {
1511 WARN_ON(start
> offset
+ len
);
1513 end
= start
+ frag_iter
->len
;
1514 if ((copy
= end
- offset
) > 0) {
1517 if (skb_copy_bits(frag_iter
, offset
- start
, to
, copy
))
1519 if ((len
-= copy
) == 0)
1533 EXPORT_SYMBOL(skb_copy_bits
);
1536 * Callback from splice_to_pipe(), if we need to release some pages
1537 * at the end of the spd in case we error'ed out in filling the pipe.
1539 static void sock_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
1541 put_page(spd
->pages
[i
]);
1544 static struct page
*linear_to_page(struct page
*page
, unsigned int *len
,
1545 unsigned int *offset
,
1546 struct sk_buff
*skb
, struct sock
*sk
)
1548 struct page
*p
= sk
->sk_sndmsg_page
;
1553 p
= sk
->sk_sndmsg_page
= alloc_pages(sk
->sk_allocation
, 0);
1557 off
= sk
->sk_sndmsg_off
= 0;
1558 /* hold one ref to this page until it's full */
1562 /* If we are the only user of the page, we can reset offset */
1563 if (page_count(p
) == 1)
1564 sk
->sk_sndmsg_off
= 0;
1565 off
= sk
->sk_sndmsg_off
;
1566 mlen
= PAGE_SIZE
- off
;
1567 if (mlen
< 64 && mlen
< *len
) {
1572 *len
= min_t(unsigned int, *len
, mlen
);
1575 memcpy(page_address(p
) + off
, page_address(page
) + *offset
, *len
);
1576 sk
->sk_sndmsg_off
+= *len
;
1582 static bool spd_can_coalesce(const struct splice_pipe_desc
*spd
,
1584 unsigned int offset
)
1586 return spd
->nr_pages
&&
1587 spd
->pages
[spd
->nr_pages
- 1] == page
&&
1588 (spd
->partial
[spd
->nr_pages
- 1].offset
+
1589 spd
->partial
[spd
->nr_pages
- 1].len
== offset
);
1593 * Fill page/offset/length into spd, if it can hold more pages.
1595 static bool spd_fill_page(struct splice_pipe_desc
*spd
,
1596 struct pipe_inode_info
*pipe
, struct page
*page
,
1597 unsigned int *len
, unsigned int offset
,
1598 struct sk_buff
*skb
, bool linear
,
1601 if (unlikely(spd
->nr_pages
== MAX_SKB_FRAGS
))
1605 page
= linear_to_page(page
, len
, &offset
, skb
, sk
);
1609 if (spd_can_coalesce(spd
, page
, offset
)) {
1610 spd
->partial
[spd
->nr_pages
- 1].len
+= *len
;
1614 spd
->pages
[spd
->nr_pages
] = page
;
1615 spd
->partial
[spd
->nr_pages
].len
= *len
;
1616 spd
->partial
[spd
->nr_pages
].offset
= offset
;
1622 static inline void __segment_seek(struct page
**page
, unsigned int *poff
,
1623 unsigned int *plen
, unsigned int off
)
1628 n
= *poff
/ PAGE_SIZE
;
1630 *page
= nth_page(*page
, n
);
1632 *poff
= *poff
% PAGE_SIZE
;
1636 static bool __splice_segment(struct page
*page
, unsigned int poff
,
1637 unsigned int plen
, unsigned int *off
,
1638 unsigned int *len
, struct sk_buff
*skb
,
1639 struct splice_pipe_desc
*spd
, bool linear
,
1641 struct pipe_inode_info
*pipe
)
1646 /* skip this segment if already processed */
1652 /* ignore any bits we already processed */
1654 __segment_seek(&page
, &poff
, &plen
, *off
);
1659 unsigned int flen
= min(*len
, plen
);
1661 /* the linear region may spread across several pages */
1662 flen
= min_t(unsigned int, flen
, PAGE_SIZE
- poff
);
1664 if (spd_fill_page(spd
, pipe
, page
, &flen
, poff
, skb
, linear
, sk
))
1667 __segment_seek(&page
, &poff
, &plen
, flen
);
1670 } while (*len
&& plen
);
1676 * Map linear and fragment data from the skb to spd. It reports true if the
1677 * pipe is full or if we already spliced the requested length.
1679 static bool __skb_splice_bits(struct sk_buff
*skb
, struct pipe_inode_info
*pipe
,
1680 unsigned int *offset
, unsigned int *len
,
1681 struct splice_pipe_desc
*spd
, struct sock
*sk
)
1685 /* map the linear part :
1686 * If skb->head_frag is set, this 'linear' part is backed by a
1687 * fragment, and if the head is not shared with any clones then
1688 * we can avoid a copy since we own the head portion of this page.
1690 if (__splice_segment(virt_to_page(skb
->data
),
1691 (unsigned long) skb
->data
& (PAGE_SIZE
- 1),
1693 offset
, len
, skb
, spd
,
1694 skb_head_is_locked(skb
),
1699 * then map the fragments
1701 for (seg
= 0; seg
< skb_shinfo(skb
)->nr_frags
; seg
++) {
1702 const skb_frag_t
*f
= &skb_shinfo(skb
)->frags
[seg
];
1704 if (__splice_segment(skb_frag_page(f
),
1705 f
->page_offset
, skb_frag_size(f
),
1706 offset
, len
, skb
, spd
, false, sk
, pipe
))
1714 * Map data from the skb to a pipe. Should handle both the linear part,
1715 * the fragments, and the frag list. It does NOT handle frag lists within
1716 * the frag list, if such a thing exists. We'd probably need to recurse to
1717 * handle that cleanly.
1719 int skb_splice_bits(struct sk_buff
*skb
, unsigned int offset
,
1720 struct pipe_inode_info
*pipe
, unsigned int tlen
,
1723 struct partial_page partial
[MAX_SKB_FRAGS
];
1724 struct page
*pages
[MAX_SKB_FRAGS
];
1725 struct splice_pipe_desc spd
= {
1729 .ops
= &sock_pipe_buf_ops
,
1730 .spd_release
= sock_spd_release
,
1732 struct sk_buff
*frag_iter
;
1733 struct sock
*sk
= skb
->sk
;
1737 * __skb_splice_bits() only fails if the output has no room left,
1738 * so no point in going over the frag_list for the error case.
1740 if (__skb_splice_bits(skb
, pipe
, &offset
, &tlen
, &spd
, sk
))
1746 * now see if we have a frag_list to map
1748 skb_walk_frags(skb
, frag_iter
) {
1751 if (__skb_splice_bits(frag_iter
, pipe
, &offset
, &tlen
, &spd
, sk
))
1758 * Drop the socket lock, otherwise we have reverse
1759 * locking dependencies between sk_lock and i_mutex
1760 * here as compared to sendfile(). We enter here
1761 * with the socket lock held, and splice_to_pipe() will
1762 * grab the pipe inode lock. For sendfile() emulation,
1763 * we call into ->sendpage() with the i_mutex lock held
1764 * and networking will grab the socket lock.
1767 ret
= splice_to_pipe(pipe
, &spd
);
1775 * skb_store_bits - store bits from kernel buffer to skb
1776 * @skb: destination buffer
1777 * @offset: offset in destination
1778 * @from: source buffer
1779 * @len: number of bytes to copy
1781 * Copy the specified number of bytes from the source buffer to the
1782 * destination skb. This function handles all the messy bits of
1783 * traversing fragment lists and such.
1786 int skb_store_bits(struct sk_buff
*skb
, int offset
, const void *from
, int len
)
1788 int start
= skb_headlen(skb
);
1789 struct sk_buff
*frag_iter
;
1792 if (offset
> (int)skb
->len
- len
)
1795 if ((copy
= start
- offset
) > 0) {
1798 skb_copy_to_linear_data_offset(skb
, offset
, from
, copy
);
1799 if ((len
-= copy
) == 0)
1805 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1806 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1809 WARN_ON(start
> offset
+ len
);
1811 end
= start
+ skb_frag_size(frag
);
1812 if ((copy
= end
- offset
) > 0) {
1818 vaddr
= kmap_atomic(skb_frag_page(frag
));
1819 memcpy(vaddr
+ frag
->page_offset
+ offset
- start
,
1821 kunmap_atomic(vaddr
);
1823 if ((len
-= copy
) == 0)
1831 skb_walk_frags(skb
, frag_iter
) {
1834 WARN_ON(start
> offset
+ len
);
1836 end
= start
+ frag_iter
->len
;
1837 if ((copy
= end
- offset
) > 0) {
1840 if (skb_store_bits(frag_iter
, offset
- start
,
1843 if ((len
-= copy
) == 0)
1856 EXPORT_SYMBOL(skb_store_bits
);
1858 /* Checksum skb data. */
1860 __wsum
skb_checksum(const struct sk_buff
*skb
, int offset
,
1861 int len
, __wsum csum
)
1863 int start
= skb_headlen(skb
);
1864 int i
, copy
= start
- offset
;
1865 struct sk_buff
*frag_iter
;
1868 /* Checksum header. */
1872 csum
= csum_partial(skb
->data
+ offset
, copy
, csum
);
1873 if ((len
-= copy
) == 0)
1879 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1881 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1883 WARN_ON(start
> offset
+ len
);
1885 end
= start
+ skb_frag_size(frag
);
1886 if ((copy
= end
- offset
) > 0) {
1892 vaddr
= kmap_atomic(skb_frag_page(frag
));
1893 csum2
= csum_partial(vaddr
+ frag
->page_offset
+
1894 offset
- start
, copy
, 0);
1895 kunmap_atomic(vaddr
);
1896 csum
= csum_block_add(csum
, csum2
, pos
);
1905 skb_walk_frags(skb
, frag_iter
) {
1908 WARN_ON(start
> offset
+ len
);
1910 end
= start
+ frag_iter
->len
;
1911 if ((copy
= end
- offset
) > 0) {
1915 csum2
= skb_checksum(frag_iter
, offset
- start
,
1917 csum
= csum_block_add(csum
, csum2
, pos
);
1918 if ((len
-= copy
) == 0)
1929 EXPORT_SYMBOL(skb_checksum
);
1931 /* Both of above in one bottle. */
1933 __wsum
skb_copy_and_csum_bits(const struct sk_buff
*skb
, int offset
,
1934 u8
*to
, int len
, __wsum csum
)
1936 int start
= skb_headlen(skb
);
1937 int i
, copy
= start
- offset
;
1938 struct sk_buff
*frag_iter
;
1945 csum
= csum_partial_copy_nocheck(skb
->data
+ offset
, to
,
1947 if ((len
-= copy
) == 0)
1954 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
1957 WARN_ON(start
> offset
+ len
);
1959 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
1960 if ((copy
= end
- offset
) > 0) {
1963 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1967 vaddr
= kmap_atomic(skb_frag_page(frag
));
1968 csum2
= csum_partial_copy_nocheck(vaddr
+
1972 kunmap_atomic(vaddr
);
1973 csum
= csum_block_add(csum
, csum2
, pos
);
1983 skb_walk_frags(skb
, frag_iter
) {
1987 WARN_ON(start
> offset
+ len
);
1989 end
= start
+ frag_iter
->len
;
1990 if ((copy
= end
- offset
) > 0) {
1993 csum2
= skb_copy_and_csum_bits(frag_iter
,
1996 csum
= csum_block_add(csum
, csum2
, pos
);
1997 if ((len
-= copy
) == 0)
2008 EXPORT_SYMBOL(skb_copy_and_csum_bits
);
2010 void skb_copy_and_csum_dev(const struct sk_buff
*skb
, u8
*to
)
2015 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
2016 csstart
= skb_checksum_start_offset(skb
);
2018 csstart
= skb_headlen(skb
);
2020 BUG_ON(csstart
> skb_headlen(skb
));
2022 skb_copy_from_linear_data(skb
, to
, csstart
);
2025 if (csstart
!= skb
->len
)
2026 csum
= skb_copy_and_csum_bits(skb
, csstart
, to
+ csstart
,
2027 skb
->len
- csstart
, 0);
2029 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2030 long csstuff
= csstart
+ skb
->csum_offset
;
2032 *((__sum16
*)(to
+ csstuff
)) = csum_fold(csum
);
2035 EXPORT_SYMBOL(skb_copy_and_csum_dev
);
2038 * skb_dequeue - remove from the head of the queue
2039 * @list: list to dequeue from
2041 * Remove the head of the list. The list lock is taken so the function
2042 * may be used safely with other locking list functions. The head item is
2043 * returned or %NULL if the list is empty.
2046 struct sk_buff
*skb_dequeue(struct sk_buff_head
*list
)
2048 unsigned long flags
;
2049 struct sk_buff
*result
;
2051 spin_lock_irqsave(&list
->lock
, flags
);
2052 result
= __skb_dequeue(list
);
2053 spin_unlock_irqrestore(&list
->lock
, flags
);
2056 EXPORT_SYMBOL(skb_dequeue
);
2059 * skb_dequeue_tail - remove from the tail of the queue
2060 * @list: list to dequeue from
2062 * Remove the tail of the list. The list lock is taken so the function
2063 * may be used safely with other locking list functions. The tail item is
2064 * returned or %NULL if the list is empty.
2066 struct sk_buff
*skb_dequeue_tail(struct sk_buff_head
*list
)
2068 unsigned long flags
;
2069 struct sk_buff
*result
;
2071 spin_lock_irqsave(&list
->lock
, flags
);
2072 result
= __skb_dequeue_tail(list
);
2073 spin_unlock_irqrestore(&list
->lock
, flags
);
2076 EXPORT_SYMBOL(skb_dequeue_tail
);
2079 * skb_queue_purge - empty a list
2080 * @list: list to empty
2082 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2083 * the list and one reference dropped. This function takes the list
2084 * lock and is atomic with respect to other list locking functions.
2086 void skb_queue_purge(struct sk_buff_head
*list
)
2088 struct sk_buff
*skb
;
2089 while ((skb
= skb_dequeue(list
)) != NULL
)
2092 EXPORT_SYMBOL(skb_queue_purge
);
2095 * skb_queue_head - queue a buffer at the list head
2096 * @list: list to use
2097 * @newsk: buffer to queue
2099 * Queue a buffer at the start of the list. This function takes the
2100 * list lock and can be used safely with other locking &sk_buff functions
2103 * A buffer cannot be placed on two lists at the same time.
2105 void skb_queue_head(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2107 unsigned long flags
;
2109 spin_lock_irqsave(&list
->lock
, flags
);
2110 __skb_queue_head(list
, newsk
);
2111 spin_unlock_irqrestore(&list
->lock
, flags
);
2113 EXPORT_SYMBOL(skb_queue_head
);
2116 * skb_queue_tail - queue a buffer at the list tail
2117 * @list: list to use
2118 * @newsk: buffer to queue
2120 * Queue a buffer at the tail of the list. This function takes the
2121 * list lock and can be used safely with other locking &sk_buff functions
2124 * A buffer cannot be placed on two lists at the same time.
2126 void skb_queue_tail(struct sk_buff_head
*list
, struct sk_buff
*newsk
)
2128 unsigned long flags
;
2130 spin_lock_irqsave(&list
->lock
, flags
);
2131 __skb_queue_tail(list
, newsk
);
2132 spin_unlock_irqrestore(&list
->lock
, flags
);
2134 EXPORT_SYMBOL(skb_queue_tail
);
2137 * skb_unlink - remove a buffer from a list
2138 * @skb: buffer to remove
2139 * @list: list to use
2141 * Remove a packet from a list. The list locks are taken and this
2142 * function is atomic with respect to other list locked calls
2144 * You must know what list the SKB is on.
2146 void skb_unlink(struct sk_buff
*skb
, struct sk_buff_head
*list
)
2148 unsigned long flags
;
2150 spin_lock_irqsave(&list
->lock
, flags
);
2151 __skb_unlink(skb
, list
);
2152 spin_unlock_irqrestore(&list
->lock
, flags
);
2154 EXPORT_SYMBOL(skb_unlink
);
2157 * skb_append - append a buffer
2158 * @old: buffer to insert after
2159 * @newsk: buffer to insert
2160 * @list: list to use
2162 * Place a packet after a given packet in a list. The list locks are taken
2163 * and this function is atomic with respect to other list locked calls.
2164 * A buffer cannot be placed on two lists at the same time.
2166 void skb_append(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2168 unsigned long flags
;
2170 spin_lock_irqsave(&list
->lock
, flags
);
2171 __skb_queue_after(list
, old
, newsk
);
2172 spin_unlock_irqrestore(&list
->lock
, flags
);
2174 EXPORT_SYMBOL(skb_append
);
2177 * skb_insert - insert a buffer
2178 * @old: buffer to insert before
2179 * @newsk: buffer to insert
2180 * @list: list to use
2182 * Place a packet before a given packet in a list. The list locks are
2183 * taken and this function is atomic with respect to other list locked
2186 * A buffer cannot be placed on two lists at the same time.
2188 void skb_insert(struct sk_buff
*old
, struct sk_buff
*newsk
, struct sk_buff_head
*list
)
2190 unsigned long flags
;
2192 spin_lock_irqsave(&list
->lock
, flags
);
2193 __skb_insert(newsk
, old
->prev
, old
, list
);
2194 spin_unlock_irqrestore(&list
->lock
, flags
);
2196 EXPORT_SYMBOL(skb_insert
);
2198 static inline void skb_split_inside_header(struct sk_buff
*skb
,
2199 struct sk_buff
* skb1
,
2200 const u32 len
, const int pos
)
2204 skb_copy_from_linear_data_offset(skb
, len
, skb_put(skb1
, pos
- len
),
2206 /* And move data appendix as is. */
2207 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
2208 skb_shinfo(skb1
)->frags
[i
] = skb_shinfo(skb
)->frags
[i
];
2210 skb_shinfo(skb1
)->nr_frags
= skb_shinfo(skb
)->nr_frags
;
2211 skb_shinfo(skb
)->nr_frags
= 0;
2212 skb1
->data_len
= skb
->data_len
;
2213 skb1
->len
+= skb1
->data_len
;
2216 skb_set_tail_pointer(skb
, len
);
2219 static inline void skb_split_no_header(struct sk_buff
*skb
,
2220 struct sk_buff
* skb1
,
2221 const u32 len
, int pos
)
2224 const int nfrags
= skb_shinfo(skb
)->nr_frags
;
2226 skb_shinfo(skb
)->nr_frags
= 0;
2227 skb1
->len
= skb1
->data_len
= skb
->len
- len
;
2229 skb
->data_len
= len
- pos
;
2231 for (i
= 0; i
< nfrags
; i
++) {
2232 int size
= skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
2234 if (pos
+ size
> len
) {
2235 skb_shinfo(skb1
)->frags
[k
] = skb_shinfo(skb
)->frags
[i
];
2239 * We have two variants in this case:
2240 * 1. Move all the frag to the second
2241 * part, if it is possible. F.e.
2242 * this approach is mandatory for TUX,
2243 * where splitting is expensive.
2244 * 2. Split is accurately. We make this.
2246 skb_frag_ref(skb
, i
);
2247 skb_shinfo(skb1
)->frags
[0].page_offset
+= len
- pos
;
2248 skb_frag_size_sub(&skb_shinfo(skb1
)->frags
[0], len
- pos
);
2249 skb_frag_size_set(&skb_shinfo(skb
)->frags
[i
], len
- pos
);
2250 skb_shinfo(skb
)->nr_frags
++;
2254 skb_shinfo(skb
)->nr_frags
++;
2257 skb_shinfo(skb1
)->nr_frags
= k
;
2261 * skb_split - Split fragmented skb to two parts at length len.
2262 * @skb: the buffer to split
2263 * @skb1: the buffer to receive the second part
2264 * @len: new length for skb
2266 void skb_split(struct sk_buff
*skb
, struct sk_buff
*skb1
, const u32 len
)
2268 int pos
= skb_headlen(skb
);
2270 if (len
< pos
) /* Split line is inside header. */
2271 skb_split_inside_header(skb
, skb1
, len
, pos
);
2272 else /* Second chunk has no header, nothing to copy. */
2273 skb_split_no_header(skb
, skb1
, len
, pos
);
2275 EXPORT_SYMBOL(skb_split
);
2277 /* Shifting from/to a cloned skb is a no-go.
2279 * Caller cannot keep skb_shinfo related pointers past calling here!
2281 static int skb_prepare_for_shift(struct sk_buff
*skb
)
2283 return skb_cloned(skb
) && pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
2287 * skb_shift - Shifts paged data partially from skb to another
2288 * @tgt: buffer into which tail data gets added
2289 * @skb: buffer from which the paged data comes from
2290 * @shiftlen: shift up to this many bytes
2292 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2293 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2294 * It's up to caller to free skb if everything was shifted.
2296 * If @tgt runs out of frags, the whole operation is aborted.
2298 * Skb cannot include anything else but paged data while tgt is allowed
2299 * to have non-paged data as well.
2301 * TODO: full sized shift could be optimized but that would need
2302 * specialized skb free'er to handle frags without up-to-date nr_frags.
2304 int skb_shift(struct sk_buff
*tgt
, struct sk_buff
*skb
, int shiftlen
)
2306 int from
, to
, merge
, todo
;
2307 struct skb_frag_struct
*fragfrom
, *fragto
;
2309 BUG_ON(shiftlen
> skb
->len
);
2310 BUG_ON(skb_headlen(skb
)); /* Would corrupt stream */
2314 to
= skb_shinfo(tgt
)->nr_frags
;
2315 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2317 /* Actual merge is delayed until the point when we know we can
2318 * commit all, so that we don't have to undo partial changes
2321 !skb_can_coalesce(tgt
, to
, skb_frag_page(fragfrom
),
2322 fragfrom
->page_offset
)) {
2327 todo
-= skb_frag_size(fragfrom
);
2329 if (skb_prepare_for_shift(skb
) ||
2330 skb_prepare_for_shift(tgt
))
2333 /* All previous frag pointers might be stale! */
2334 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2335 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2337 skb_frag_size_add(fragto
, shiftlen
);
2338 skb_frag_size_sub(fragfrom
, shiftlen
);
2339 fragfrom
->page_offset
+= shiftlen
;
2347 /* Skip full, not-fitting skb to avoid expensive operations */
2348 if ((shiftlen
== skb
->len
) &&
2349 (skb_shinfo(skb
)->nr_frags
- from
) > (MAX_SKB_FRAGS
- to
))
2352 if (skb_prepare_for_shift(skb
) || skb_prepare_for_shift(tgt
))
2355 while ((todo
> 0) && (from
< skb_shinfo(skb
)->nr_frags
)) {
2356 if (to
== MAX_SKB_FRAGS
)
2359 fragfrom
= &skb_shinfo(skb
)->frags
[from
];
2360 fragto
= &skb_shinfo(tgt
)->frags
[to
];
2362 if (todo
>= skb_frag_size(fragfrom
)) {
2363 *fragto
= *fragfrom
;
2364 todo
-= skb_frag_size(fragfrom
);
2369 __skb_frag_ref(fragfrom
);
2370 fragto
->page
= fragfrom
->page
;
2371 fragto
->page_offset
= fragfrom
->page_offset
;
2372 skb_frag_size_set(fragto
, todo
);
2374 fragfrom
->page_offset
+= todo
;
2375 skb_frag_size_sub(fragfrom
, todo
);
2383 /* Ready to "commit" this state change to tgt */
2384 skb_shinfo(tgt
)->nr_frags
= to
;
2387 fragfrom
= &skb_shinfo(skb
)->frags
[0];
2388 fragto
= &skb_shinfo(tgt
)->frags
[merge
];
2390 skb_frag_size_add(fragto
, skb_frag_size(fragfrom
));
2391 __skb_frag_unref(fragfrom
);
2394 /* Reposition in the original skb */
2396 while (from
< skb_shinfo(skb
)->nr_frags
)
2397 skb_shinfo(skb
)->frags
[to
++] = skb_shinfo(skb
)->frags
[from
++];
2398 skb_shinfo(skb
)->nr_frags
= to
;
2400 BUG_ON(todo
> 0 && !skb_shinfo(skb
)->nr_frags
);
2403 /* Most likely the tgt won't ever need its checksum anymore, skb on
2404 * the other hand might need it if it needs to be resent
2406 tgt
->ip_summed
= CHECKSUM_PARTIAL
;
2407 skb
->ip_summed
= CHECKSUM_PARTIAL
;
2409 /* Yak, is it really working this way? Some helper please? */
2410 skb
->len
-= shiftlen
;
2411 skb
->data_len
-= shiftlen
;
2412 skb
->truesize
-= shiftlen
;
2413 tgt
->len
+= shiftlen
;
2414 tgt
->data_len
+= shiftlen
;
2415 tgt
->truesize
+= shiftlen
;
2421 * skb_prepare_seq_read - Prepare a sequential read of skb data
2422 * @skb: the buffer to read
2423 * @from: lower offset of data to be read
2424 * @to: upper offset of data to be read
2425 * @st: state variable
2427 * Initializes the specified state variable. Must be called before
2428 * invoking skb_seq_read() for the first time.
2430 void skb_prepare_seq_read(struct sk_buff
*skb
, unsigned int from
,
2431 unsigned int to
, struct skb_seq_state
*st
)
2433 st
->lower_offset
= from
;
2434 st
->upper_offset
= to
;
2435 st
->root_skb
= st
->cur_skb
= skb
;
2436 st
->frag_idx
= st
->stepped_offset
= 0;
2437 st
->frag_data
= NULL
;
2439 EXPORT_SYMBOL(skb_prepare_seq_read
);
2442 * skb_seq_read - Sequentially read skb data
2443 * @consumed: number of bytes consumed by the caller so far
2444 * @data: destination pointer for data to be returned
2445 * @st: state variable
2447 * Reads a block of skb data at &consumed relative to the
2448 * lower offset specified to skb_prepare_seq_read(). Assigns
2449 * the head of the data block to &data and returns the length
2450 * of the block or 0 if the end of the skb data or the upper
2451 * offset has been reached.
2453 * The caller is not required to consume all of the data
2454 * returned, i.e. &consumed is typically set to the number
2455 * of bytes already consumed and the next call to
2456 * skb_seq_read() will return the remaining part of the block.
2458 * Note 1: The size of each block of data returned can be arbitrary,
2459 * this limitation is the cost for zerocopy seqeuental
2460 * reads of potentially non linear data.
2462 * Note 2: Fragment lists within fragments are not implemented
2463 * at the moment, state->root_skb could be replaced with
2464 * a stack for this purpose.
2466 unsigned int skb_seq_read(unsigned int consumed
, const u8
**data
,
2467 struct skb_seq_state
*st
)
2469 unsigned int block_limit
, abs_offset
= consumed
+ st
->lower_offset
;
2472 if (unlikely(abs_offset
>= st
->upper_offset
))
2476 block_limit
= skb_headlen(st
->cur_skb
) + st
->stepped_offset
;
2478 if (abs_offset
< block_limit
&& !st
->frag_data
) {
2479 *data
= st
->cur_skb
->data
+ (abs_offset
- st
->stepped_offset
);
2480 return block_limit
- abs_offset
;
2483 if (st
->frag_idx
== 0 && !st
->frag_data
)
2484 st
->stepped_offset
+= skb_headlen(st
->cur_skb
);
2486 while (st
->frag_idx
< skb_shinfo(st
->cur_skb
)->nr_frags
) {
2487 frag
= &skb_shinfo(st
->cur_skb
)->frags
[st
->frag_idx
];
2488 block_limit
= skb_frag_size(frag
) + st
->stepped_offset
;
2490 if (abs_offset
< block_limit
) {
2492 st
->frag_data
= kmap_atomic(skb_frag_page(frag
));
2494 *data
= (u8
*) st
->frag_data
+ frag
->page_offset
+
2495 (abs_offset
- st
->stepped_offset
);
2497 return block_limit
- abs_offset
;
2500 if (st
->frag_data
) {
2501 kunmap_atomic(st
->frag_data
);
2502 st
->frag_data
= NULL
;
2506 st
->stepped_offset
+= skb_frag_size(frag
);
2509 if (st
->frag_data
) {
2510 kunmap_atomic(st
->frag_data
);
2511 st
->frag_data
= NULL
;
2514 if (st
->root_skb
== st
->cur_skb
&& skb_has_frag_list(st
->root_skb
)) {
2515 st
->cur_skb
= skb_shinfo(st
->root_skb
)->frag_list
;
2518 } else if (st
->cur_skb
->next
) {
2519 st
->cur_skb
= st
->cur_skb
->next
;
2526 EXPORT_SYMBOL(skb_seq_read
);
2529 * skb_abort_seq_read - Abort a sequential read of skb data
2530 * @st: state variable
2532 * Must be called if skb_seq_read() was not called until it
2535 void skb_abort_seq_read(struct skb_seq_state
*st
)
2538 kunmap_atomic(st
->frag_data
);
2540 EXPORT_SYMBOL(skb_abort_seq_read
);
2542 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2544 static unsigned int skb_ts_get_next_block(unsigned int offset
, const u8
**text
,
2545 struct ts_config
*conf
,
2546 struct ts_state
*state
)
2548 return skb_seq_read(offset
, text
, TS_SKB_CB(state
));
2551 static void skb_ts_finish(struct ts_config
*conf
, struct ts_state
*state
)
2553 skb_abort_seq_read(TS_SKB_CB(state
));
2557 * skb_find_text - Find a text pattern in skb data
2558 * @skb: the buffer to look in
2559 * @from: search offset
2561 * @config: textsearch configuration
2562 * @state: uninitialized textsearch state variable
2564 * Finds a pattern in the skb data according to the specified
2565 * textsearch configuration. Use textsearch_next() to retrieve
2566 * subsequent occurrences of the pattern. Returns the offset
2567 * to the first occurrence or UINT_MAX if no match was found.
2569 unsigned int skb_find_text(struct sk_buff
*skb
, unsigned int from
,
2570 unsigned int to
, struct ts_config
*config
,
2571 struct ts_state
*state
)
2575 config
->get_next_block
= skb_ts_get_next_block
;
2576 config
->finish
= skb_ts_finish
;
2578 skb_prepare_seq_read(skb
, from
, to
, TS_SKB_CB(state
));
2580 ret
= textsearch_find(config
, state
);
2581 return (ret
<= to
- from
? ret
: UINT_MAX
);
2583 EXPORT_SYMBOL(skb_find_text
);
2586 * skb_append_datato_frags: - append the user data to a skb
2587 * @sk: sock structure
2588 * @skb: skb structure to be appened with user data.
2589 * @getfrag: call back function to be used for getting the user data
2590 * @from: pointer to user message iov
2591 * @length: length of the iov message
2593 * Description: This procedure append the user data in the fragment part
2594 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2596 int skb_append_datato_frags(struct sock
*sk
, struct sk_buff
*skb
,
2597 int (*getfrag
)(void *from
, char *to
, int offset
,
2598 int len
, int odd
, struct sk_buff
*skb
),
2599 void *from
, int length
)
2602 skb_frag_t
*frag
= NULL
;
2603 struct page
*page
= NULL
;
2609 /* Return error if we don't have space for new frag */
2610 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2611 if (frg_cnt
>= MAX_SKB_FRAGS
)
2614 /* allocate a new page for next frag */
2615 page
= alloc_pages(sk
->sk_allocation
, 0);
2617 /* If alloc_page fails just return failure and caller will
2618 * free previous allocated pages by doing kfree_skb()
2623 /* initialize the next frag */
2624 skb_fill_page_desc(skb
, frg_cnt
, page
, 0, 0);
2625 skb
->truesize
+= PAGE_SIZE
;
2626 atomic_add(PAGE_SIZE
, &sk
->sk_wmem_alloc
);
2628 /* get the new initialized frag */
2629 frg_cnt
= skb_shinfo(skb
)->nr_frags
;
2630 frag
= &skb_shinfo(skb
)->frags
[frg_cnt
- 1];
2632 /* copy the user data to page */
2633 left
= PAGE_SIZE
- frag
->page_offset
;
2634 copy
= (length
> left
)? left
: length
;
2636 ret
= getfrag(from
, skb_frag_address(frag
) + skb_frag_size(frag
),
2637 offset
, copy
, 0, skb
);
2641 /* copy was successful so update the size parameters */
2642 skb_frag_size_add(frag
, copy
);
2644 skb
->data_len
+= copy
;
2648 } while (length
> 0);
2652 EXPORT_SYMBOL(skb_append_datato_frags
);
2655 * skb_pull_rcsum - pull skb and update receive checksum
2656 * @skb: buffer to update
2657 * @len: length of data pulled
2659 * This function performs an skb_pull on the packet and updates
2660 * the CHECKSUM_COMPLETE checksum. It should be used on
2661 * receive path processing instead of skb_pull unless you know
2662 * that the checksum difference is zero (e.g., a valid IP header)
2663 * or you are setting ip_summed to CHECKSUM_NONE.
2665 unsigned char *skb_pull_rcsum(struct sk_buff
*skb
, unsigned int len
)
2667 BUG_ON(len
> skb
->len
);
2669 BUG_ON(skb
->len
< skb
->data_len
);
2670 skb_postpull_rcsum(skb
, skb
->data
, len
);
2671 return skb
->data
+= len
;
2673 EXPORT_SYMBOL_GPL(skb_pull_rcsum
);
2676 * skb_segment - Perform protocol segmentation on skb.
2677 * @skb: buffer to segment
2678 * @features: features for the output path (see dev->features)
2680 * This function performs segmentation on the given skb. It returns
2681 * a pointer to the first in a list of new skbs for the segments.
2682 * In case of error it returns ERR_PTR(err).
2684 struct sk_buff
*skb_segment(struct sk_buff
*skb
, netdev_features_t features
)
2686 struct sk_buff
*segs
= NULL
;
2687 struct sk_buff
*tail
= NULL
;
2688 struct sk_buff
*fskb
= skb_shinfo(skb
)->frag_list
;
2689 unsigned int mss
= skb_shinfo(skb
)->gso_size
;
2690 unsigned int doffset
= skb
->data
- skb_mac_header(skb
);
2691 unsigned int offset
= doffset
;
2692 unsigned int headroom
;
2694 int sg
= !!(features
& NETIF_F_SG
);
2695 int nfrags
= skb_shinfo(skb
)->nr_frags
;
2700 __skb_push(skb
, doffset
);
2701 headroom
= skb_headroom(skb
);
2702 pos
= skb_headlen(skb
);
2705 struct sk_buff
*nskb
;
2710 len
= skb
->len
- offset
;
2714 hsize
= skb_headlen(skb
) - offset
;
2717 if (hsize
> len
|| !sg
)
2720 if (!hsize
&& i
>= nfrags
) {
2721 BUG_ON(fskb
->len
!= len
);
2724 nskb
= skb_clone(fskb
, GFP_ATOMIC
);
2727 if (unlikely(!nskb
))
2730 hsize
= skb_end_offset(nskb
);
2731 if (skb_cow_head(nskb
, doffset
+ headroom
)) {
2736 nskb
->truesize
+= skb_end_offset(nskb
) - hsize
;
2737 skb_release_head_state(nskb
);
2738 __skb_push(nskb
, doffset
);
2740 nskb
= alloc_skb(hsize
+ doffset
+ headroom
,
2743 if (unlikely(!nskb
))
2746 skb_reserve(nskb
, headroom
);
2747 __skb_put(nskb
, doffset
);
2756 __copy_skb_header(nskb
, skb
);
2757 nskb
->mac_len
= skb
->mac_len
;
2759 /* nskb and skb might have different headroom */
2760 if (nskb
->ip_summed
== CHECKSUM_PARTIAL
)
2761 nskb
->csum_start
+= skb_headroom(nskb
) - headroom
;
2763 skb_reset_mac_header(nskb
);
2764 skb_set_network_header(nskb
, skb
->mac_len
);
2765 nskb
->transport_header
= (nskb
->network_header
+
2766 skb_network_header_len(skb
));
2767 skb_copy_from_linear_data(skb
, nskb
->data
, doffset
);
2769 if (fskb
!= skb_shinfo(skb
)->frag_list
)
2773 nskb
->ip_summed
= CHECKSUM_NONE
;
2774 nskb
->csum
= skb_copy_and_csum_bits(skb
, offset
,
2780 frag
= skb_shinfo(nskb
)->frags
;
2782 skb_copy_from_linear_data_offset(skb
, offset
,
2783 skb_put(nskb
, hsize
), hsize
);
2785 while (pos
< offset
+ len
&& i
< nfrags
) {
2786 *frag
= skb_shinfo(skb
)->frags
[i
];
2787 __skb_frag_ref(frag
);
2788 size
= skb_frag_size(frag
);
2791 frag
->page_offset
+= offset
- pos
;
2792 skb_frag_size_sub(frag
, offset
- pos
);
2795 skb_shinfo(nskb
)->nr_frags
++;
2797 if (pos
+ size
<= offset
+ len
) {
2801 skb_frag_size_sub(frag
, pos
+ size
- (offset
+ len
));
2808 if (pos
< offset
+ len
) {
2809 struct sk_buff
*fskb2
= fskb
;
2811 BUG_ON(pos
+ fskb
->len
!= offset
+ len
);
2817 fskb2
= skb_clone(fskb2
, GFP_ATOMIC
);
2823 SKB_FRAG_ASSERT(nskb
);
2824 skb_shinfo(nskb
)->frag_list
= fskb2
;
2828 nskb
->data_len
= len
- hsize
;
2829 nskb
->len
+= nskb
->data_len
;
2830 nskb
->truesize
+= nskb
->data_len
;
2831 } while ((offset
+= len
) < skb
->len
);
2836 while ((skb
= segs
)) {
2840 return ERR_PTR(err
);
2842 EXPORT_SYMBOL_GPL(skb_segment
);
2844 int skb_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2846 struct sk_buff
*p
= *head
;
2847 struct sk_buff
*nskb
;
2848 struct skb_shared_info
*skbinfo
= skb_shinfo(skb
);
2849 struct skb_shared_info
*pinfo
= skb_shinfo(p
);
2850 unsigned int headroom
;
2851 unsigned int len
= skb_gro_len(skb
);
2852 unsigned int offset
= skb_gro_offset(skb
);
2853 unsigned int headlen
= skb_headlen(skb
);
2854 unsigned int delta_truesize
;
2856 if (p
->len
+ len
>= 65536)
2859 if (pinfo
->frag_list
)
2861 else if (headlen
<= offset
) {
2864 int i
= skbinfo
->nr_frags
;
2865 int nr_frags
= pinfo
->nr_frags
+ i
;
2869 if (nr_frags
> MAX_SKB_FRAGS
)
2872 pinfo
->nr_frags
= nr_frags
;
2873 skbinfo
->nr_frags
= 0;
2875 frag
= pinfo
->frags
+ nr_frags
;
2876 frag2
= skbinfo
->frags
+ i
;
2881 frag
->page_offset
+= offset
;
2882 skb_frag_size_sub(frag
, offset
);
2884 /* all fragments truesize : remove (head size + sk_buff) */
2885 delta_truesize
= skb
->truesize
-
2886 SKB_TRUESIZE(skb_end_offset(skb
));
2888 skb
->truesize
-= skb
->data_len
;
2889 skb
->len
-= skb
->data_len
;
2892 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE
;
2894 } else if (skb
->head_frag
) {
2895 int nr_frags
= pinfo
->nr_frags
;
2896 skb_frag_t
*frag
= pinfo
->frags
+ nr_frags
;
2897 struct page
*page
= virt_to_head_page(skb
->head
);
2898 unsigned int first_size
= headlen
- offset
;
2899 unsigned int first_offset
;
2901 if (nr_frags
+ 1 + skbinfo
->nr_frags
> MAX_SKB_FRAGS
)
2904 first_offset
= skb
->data
-
2905 (unsigned char *)page_address(page
) +
2908 pinfo
->nr_frags
= nr_frags
+ 1 + skbinfo
->nr_frags
;
2910 frag
->page
.p
= page
;
2911 frag
->page_offset
= first_offset
;
2912 skb_frag_size_set(frag
, first_size
);
2914 memcpy(frag
+ 1, skbinfo
->frags
, sizeof(*frag
) * skbinfo
->nr_frags
);
2915 /* We dont need to clear skbinfo->nr_frags here */
2917 delta_truesize
= skb
->truesize
- SKB_DATA_ALIGN(sizeof(struct sk_buff
));
2918 NAPI_GRO_CB(skb
)->free
= NAPI_GRO_FREE_STOLEN_HEAD
;
2920 } else if (skb_gro_len(p
) != pinfo
->gso_size
)
2923 headroom
= skb_headroom(p
);
2924 nskb
= alloc_skb(headroom
+ skb_gro_offset(p
), GFP_ATOMIC
);
2925 if (unlikely(!nskb
))
2928 __copy_skb_header(nskb
, p
);
2929 nskb
->mac_len
= p
->mac_len
;
2931 skb_reserve(nskb
, headroom
);
2932 __skb_put(nskb
, skb_gro_offset(p
));
2934 skb_set_mac_header(nskb
, skb_mac_header(p
) - p
->data
);
2935 skb_set_network_header(nskb
, skb_network_offset(p
));
2936 skb_set_transport_header(nskb
, skb_transport_offset(p
));
2938 __skb_pull(p
, skb_gro_offset(p
));
2939 memcpy(skb_mac_header(nskb
), skb_mac_header(p
),
2940 p
->data
- skb_mac_header(p
));
2942 *NAPI_GRO_CB(nskb
) = *NAPI_GRO_CB(p
);
2943 skb_shinfo(nskb
)->frag_list
= p
;
2944 skb_shinfo(nskb
)->gso_size
= pinfo
->gso_size
;
2945 pinfo
->gso_size
= 0;
2946 skb_header_release(p
);
2949 nskb
->data_len
+= p
->len
;
2950 nskb
->truesize
+= p
->truesize
;
2951 nskb
->len
+= p
->len
;
2954 nskb
->next
= p
->next
;
2960 delta_truesize
= skb
->truesize
;
2961 if (offset
> headlen
) {
2962 unsigned int eat
= offset
- headlen
;
2964 skbinfo
->frags
[0].page_offset
+= eat
;
2965 skb_frag_size_sub(&skbinfo
->frags
[0], eat
);
2966 skb
->data_len
-= eat
;
2971 __skb_pull(skb
, offset
);
2973 p
->prev
->next
= skb
;
2975 skb_header_release(skb
);
2978 NAPI_GRO_CB(p
)->count
++;
2980 p
->truesize
+= delta_truesize
;
2983 NAPI_GRO_CB(skb
)->same_flow
= 1;
2986 EXPORT_SYMBOL_GPL(skb_gro_receive
);
2988 void __init
skb_init(void)
2990 skbuff_head_cache
= kmem_cache_create("skbuff_head_cache",
2991 sizeof(struct sk_buff
),
2993 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
2995 skbuff_fclone_cache
= kmem_cache_create("skbuff_fclone_cache",
2996 (2*sizeof(struct sk_buff
)) +
2999 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
,
3004 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3005 * @skb: Socket buffer containing the buffers to be mapped
3006 * @sg: The scatter-gather list to map into
3007 * @offset: The offset into the buffer's contents to start mapping
3008 * @len: Length of buffer space to be mapped
3010 * Fill the specified scatter-gather list with mappings/pointers into a
3011 * region of the buffer space attached to a socket buffer.
3014 __skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3016 int start
= skb_headlen(skb
);
3017 int i
, copy
= start
- offset
;
3018 struct sk_buff
*frag_iter
;
3024 sg_set_buf(sg
, skb
->data
+ offset
, copy
);
3026 if ((len
-= copy
) == 0)
3031 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
3034 WARN_ON(start
> offset
+ len
);
3036 end
= start
+ skb_frag_size(&skb_shinfo(skb
)->frags
[i
]);
3037 if ((copy
= end
- offset
) > 0) {
3038 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
3042 sg_set_page(&sg
[elt
], skb_frag_page(frag
), copy
,
3043 frag
->page_offset
+offset
-start
);
3052 skb_walk_frags(skb
, frag_iter
) {
3055 WARN_ON(start
> offset
+ len
);
3057 end
= start
+ frag_iter
->len
;
3058 if ((copy
= end
- offset
) > 0) {
3061 elt
+= __skb_to_sgvec(frag_iter
, sg
+elt
, offset
- start
,
3063 if ((len
-= copy
) == 0)
3073 int skb_to_sgvec(struct sk_buff
*skb
, struct scatterlist
*sg
, int offset
, int len
)
3075 int nsg
= __skb_to_sgvec(skb
, sg
, offset
, len
);
3077 sg_mark_end(&sg
[nsg
- 1]);
3081 EXPORT_SYMBOL_GPL(skb_to_sgvec
);
3084 * skb_cow_data - Check that a socket buffer's data buffers are writable
3085 * @skb: The socket buffer to check.
3086 * @tailbits: Amount of trailing space to be added
3087 * @trailer: Returned pointer to the skb where the @tailbits space begins
3089 * Make sure that the data buffers attached to a socket buffer are
3090 * writable. If they are not, private copies are made of the data buffers
3091 * and the socket buffer is set to use these instead.
3093 * If @tailbits is given, make sure that there is space to write @tailbits
3094 * bytes of data beyond current end of socket buffer. @trailer will be
3095 * set to point to the skb in which this space begins.
3097 * The number of scatterlist elements required to completely map the
3098 * COW'd and extended socket buffer will be returned.
3100 int skb_cow_data(struct sk_buff
*skb
, int tailbits
, struct sk_buff
**trailer
)
3104 struct sk_buff
*skb1
, **skb_p
;
3106 /* If skb is cloned or its head is paged, reallocate
3107 * head pulling out all the pages (pages are considered not writable
3108 * at the moment even if they are anonymous).
3110 if ((skb_cloned(skb
) || skb_shinfo(skb
)->nr_frags
) &&
3111 __pskb_pull_tail(skb
, skb_pagelen(skb
)-skb_headlen(skb
)) == NULL
)
3114 /* Easy case. Most of packets will go this way. */
3115 if (!skb_has_frag_list(skb
)) {
3116 /* A little of trouble, not enough of space for trailer.
3117 * This should not happen, when stack is tuned to generate
3118 * good frames. OK, on miss we reallocate and reserve even more
3119 * space, 128 bytes is fair. */
3121 if (skb_tailroom(skb
) < tailbits
&&
3122 pskb_expand_head(skb
, 0, tailbits
-skb_tailroom(skb
)+128, GFP_ATOMIC
))
3130 /* Misery. We are in troubles, going to mincer fragments... */
3133 skb_p
= &skb_shinfo(skb
)->frag_list
;
3136 while ((skb1
= *skb_p
) != NULL
) {
3139 /* The fragment is partially pulled by someone,
3140 * this can happen on input. Copy it and everything
3143 if (skb_shared(skb1
))
3146 /* If the skb is the last, worry about trailer. */
3148 if (skb1
->next
== NULL
&& tailbits
) {
3149 if (skb_shinfo(skb1
)->nr_frags
||
3150 skb_has_frag_list(skb1
) ||
3151 skb_tailroom(skb1
) < tailbits
)
3152 ntail
= tailbits
+ 128;
3158 skb_shinfo(skb1
)->nr_frags
||
3159 skb_has_frag_list(skb1
)) {
3160 struct sk_buff
*skb2
;
3162 /* Fuck, we are miserable poor guys... */
3164 skb2
= skb_copy(skb1
, GFP_ATOMIC
);
3166 skb2
= skb_copy_expand(skb1
,
3170 if (unlikely(skb2
== NULL
))
3174 skb_set_owner_w(skb2
, skb1
->sk
);
3176 /* Looking around. Are we still alive?
3177 * OK, link new skb, drop old one */
3179 skb2
->next
= skb1
->next
;
3186 skb_p
= &skb1
->next
;
3191 EXPORT_SYMBOL_GPL(skb_cow_data
);
3193 static void sock_rmem_free(struct sk_buff
*skb
)
3195 struct sock
*sk
= skb
->sk
;
3197 atomic_sub(skb
->truesize
, &sk
->sk_rmem_alloc
);
3201 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3203 int sock_queue_err_skb(struct sock
*sk
, struct sk_buff
*skb
)
3207 if (atomic_read(&sk
->sk_rmem_alloc
) + skb
->truesize
>=
3208 (unsigned int)sk
->sk_rcvbuf
)
3213 skb
->destructor
= sock_rmem_free
;
3214 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
3216 /* before exiting rcu section, make sure dst is refcounted */
3219 skb_queue_tail(&sk
->sk_error_queue
, skb
);
3220 if (!sock_flag(sk
, SOCK_DEAD
))
3221 sk
->sk_data_ready(sk
, len
);
3224 EXPORT_SYMBOL(sock_queue_err_skb
);
3226 void skb_tstamp_tx(struct sk_buff
*orig_skb
,
3227 struct skb_shared_hwtstamps
*hwtstamps
)
3229 struct sock
*sk
= orig_skb
->sk
;
3230 struct sock_exterr_skb
*serr
;
3231 struct sk_buff
*skb
;
3237 skb
= skb_clone(orig_skb
, GFP_ATOMIC
);
3242 *skb_hwtstamps(skb
) =
3246 * no hardware time stamps available,
3247 * so keep the shared tx_flags and only
3248 * store software time stamp
3250 skb
->tstamp
= ktime_get_real();
3253 serr
= SKB_EXT_ERR(skb
);
3254 memset(serr
, 0, sizeof(*serr
));
3255 serr
->ee
.ee_errno
= ENOMSG
;
3256 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TIMESTAMPING
;
3258 err
= sock_queue_err_skb(sk
, skb
);
3263 EXPORT_SYMBOL_GPL(skb_tstamp_tx
);
3265 void skb_complete_wifi_ack(struct sk_buff
*skb
, bool acked
)
3267 struct sock
*sk
= skb
->sk
;
3268 struct sock_exterr_skb
*serr
;
3271 skb
->wifi_acked_valid
= 1;
3272 skb
->wifi_acked
= acked
;
3274 serr
= SKB_EXT_ERR(skb
);
3275 memset(serr
, 0, sizeof(*serr
));
3276 serr
->ee
.ee_errno
= ENOMSG
;
3277 serr
->ee
.ee_origin
= SO_EE_ORIGIN_TXSTATUS
;
3279 err
= sock_queue_err_skb(sk
, skb
);
3283 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack
);
3287 * skb_partial_csum_set - set up and verify partial csum values for packet
3288 * @skb: the skb to set
3289 * @start: the number of bytes after skb->data to start checksumming.
3290 * @off: the offset from start to place the checksum.
3292 * For untrusted partially-checksummed packets, we need to make sure the values
3293 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3295 * This function checks and sets those values and skb->ip_summed: if this
3296 * returns false you should drop the packet.
3298 bool skb_partial_csum_set(struct sk_buff
*skb
, u16 start
, u16 off
)
3300 if (unlikely(start
> skb_headlen(skb
)) ||
3301 unlikely((int)start
+ off
> skb_headlen(skb
) - 2)) {
3302 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3303 start
, off
, skb_headlen(skb
));
3306 skb
->ip_summed
= CHECKSUM_PARTIAL
;
3307 skb
->csum_start
= skb_headroom(skb
) + start
;
3308 skb
->csum_offset
= off
;
3311 EXPORT_SYMBOL_GPL(skb_partial_csum_set
);
3313 void __skb_warn_lro_forwarding(const struct sk_buff
*skb
)
3315 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3318 EXPORT_SYMBOL(__skb_warn_lro_forwarding
);